Over 150 Million Users Exposed by Flaws in AI Girlfriend Apps
चैत १०, २०८२ १६:२६
Kathmandu. Millions of people using artificial intelligence (AI) for love and intimacy are now at risk of cybersecurity threats. Researchers have found security flaws in some of the most popular 'AI girlfriend' and companion (virtual friend) apps. This has revealed that hackers can access highly confidential and sensitive user data.
According to a study conducted by the cybersecurity firm Oversecured, 17 AI companion apps on the Google Play Store were found to have 14 critical and 311 high-risk security vulnerabilities. These apps have collectively been downloaded more than 150 million times.
According to the research, hackers can easily access users’ conversations (chats) from at least six AI girlfriend apps. These discussions are extremely intimate, explicit, and—most concerning of all—linked to the user's real identity.
Users who view AI as their romantic partner are disclosing incredibly private information, including details of their extramarital affairs, sexual desires, identity issues, and, in certain situations, suicidal thoughts. Ten of the 17 apps were discovered to contain these flaws, giving hackers access to chats with AI bots. This means that users are entrusting their deepest and most private secrets to a system with weaker security than that of a regular messaging app.
According to a survey conducted among Generation Z and millennial men, 43% of men consider real-life dating to be financially burdensome. On the other hand, about 20% (1 in 5) of men admitted to 'flirting' with AI, while 47% of men believe that virtual AI dating can fulfill their emotional needs. As AI continues to enter every aspect of daily life, its impact on people's personal lives has also naturally increased.
How is data being stolen?
According to researchers, a popular app with over ten million downloads was found to contain highly sensitive passwords such as ‘OpenAI API token’ and ‘Google Cloud private key’ within its code. Experts have stated that these can be extracted with basic reverse engineering techniques.
Similarly, another app has a 'cross-site scripting' (XSS) vulnerability. This allows hackers to send malicious code within private chats. Through this, they can perform actions such as viewing real-time messages, hijacking sessions, or sending fake replies within the chat. The third vulnerability is in an app known for adult content; hackers could potentially steal chat databases, photos, voice messages, and authentication tokens.
Similarly, a popular app with 50 million downloads has been found to have a weakness in its advertising software. There is a risk that a malicious advertisement could directly attack the user's stored chat database. According to researchers, most of these newly identified weaknesses remain unresolved.
Despite such sensitive data being shared, its regulation is far behind. AI girlfriend or companion apps are not classified as healthcare services or therapy products. Therefore, legal provisions like the U.S. HIPAA that protect hospital records do not apply to them.
Sergey Toshin, the founder of Oversecured, says, “These apps handle highly sensitive data just like therapy platforms. But they became so popular so quickly that the standard security measures were never incorporated into their development process.”
Most of these apps are just “wrappers.” This means they use third-party AI models like OpenAI or Google and only add their own interface and payment system. The AI provider only handles the model. However, data storage and security are the responsibility of the “wrapper” developer. Every vulnerability discovered in the research was at this "wrapper" level, which is not protected by the major AI companies.
Actions and past major data leaks
Regulatory authorities are gradually becoming aware of this issue. In 2025, Italian regulators fined the developer of the app ‘Replica’ 5 million euros for illegally processing user data, failing to provide privacy-related information, and failing to have mechanisms to protect children.
Chatbots must disclose that they are AI and refrain from manipulating vulnerable users according to regulations set forth by the European Union's AI Act. However, it has not clearly addressed how to ensure chat security.
Experts warn that the current weaknesses may only be the beginning. Earlier, when the AI girlfriend website called 'Muah.ai' was hacked, users' sexual fantasies and custom bots were made public.
Similarly, researchers had previously disclosed that the ‘Chattee Chat ’ and ‘GiMe Chat’ apps had over 43 million messages, as well as over 600,000 pictures and videos leaked from 400,000 users. The hacked data also revealed people's purchase histories, with some spending thousands of dollars on virtual relationships.
पछिल्लो अध्यावधिक: चैत १०, २०८२ १६:२६
